For normal priced desktops:

For notebooks:
AMD offers 2 decoders per processor (Trinity).
Intel offers up to 4 decoders per processors, but most sell with 2 decoders anyway.

For desktops:
AMD offers up 2,3 or 4 doecoders per processor. All options popular.
Intel offers up 2 or 4 decoders for processor. All options popular.

For workstations or high-end PCs:
Intel offers 6 decoders per processor (Core i7 39xx)
AMD is not on this market officially. You can ATX Opteron board for workstation though.

For normal priced servers:
AMD offers up to 8 decoders per processor (Opteron 62xx)
Intel offers up to 8 decoders per processor (Xeon E5).

For large memory (expansive) servers:
Intel offers 10 decoders per processor, with no AVX instructions (Xeon E7)
AMD offers 8 decoders per processor.

If you count the number of decoders you get per particular price point - this is quite even.

-------------------

When it comes to L0 caches for instructions you really need to appreciate RISC advantage. L0 cache in Snapdragon S4 is just a piece of direct-mapped cache.

http://www.anandtech.com/show/4940/qualcomm-new-snapdragon-s4-msm8960-krait-architecture/2

No need for complicated schemes to map real addresses to microops adresses or deal with instructions that decode into too much micro-ops to fit in cache. Basically low latency for free.